Are Online and O ine Prices Similar? Evidence from Large Multi … · 2019-01-17 · Are Online and...

Are Online and Offline Prices Similar?Evidence from Large Multi-Channel Retailers

By Alberto Cavallo∗

Online prices are increasingly being used for measurement andresearch applications, yet little is known about their relation toprices collected offline, where most retail transactions take place.I conduct the first large-scale comparison of prices simultaneouslycollected from the websites and physical stores of 56 large multi-channel retailers in 10 countries. I find that price levels are iden-tical about 72 percent of the time. Price changes are not synchro-nized but have similar frequencies and average sizes. These resultshave implications for National Statistical Offices, researchers us-ing online data, and anyone interested in the effect of the Interneton retail prices.

Online prices are increasingly being used for measurement and research appli-cations. Since 2008, the Billion Prices Project at MIT has been experimentingwith daily online price indexes in the US and other countries.1 National Sta-tistical Offices (NSOs) have recently started to consider the use of online datain official Consumer Price Indices (CPIs).2 In the context of academic research,online prices are being used for a wide range of topics, including the study of pricecompetition, market segmentation, price stickiness, international relative prices,and real exchange rate dynamics.3

Despite their growing appeal, an open fundamental question about online pricesis whether they are similar to the prices that can be collected offline in physicalstores. The question is important because relatively few retail transactions take

∗ Sloan School of Management, Massachusetts Institute of Technology, 77 Massachusetts Avenue,Cambridge, MA 02139, and NBER (e-mail: [email protected]). Financial support for this research wasprovided by the JFRAP at MIT Sloan and the NBER’s Economics of Digitization and Copyright Initia-tive. I greatly benefited from the comments of referees and seminar participants at the NBER/CIRW,MIT Sloan, the Ottawa Group Meeting, and the UNECE/ILO Meeting of the Group of Experts on Con-sumer Price Indices. I thank Maria Fazzolari for her outstanding work implementing and coordinatingall the data collection. I also thank the MIT and Wellesley University students that collected some ofthe offline prices: Vivian Xi, Maurizio Boano, Sibo Wang, Descartes Holland, Sabrina Lui, Suh Yoon,Holly Zhu, Sean Bingham, Elizabeth Krauthamer, Jeffrey Zhang, William Rodriguez, Wenxin Zhang,Jake Amereno, Ivy Jiang, Diya Mo, Qi Jin, Riley Quinn, Do Yeon Park, Jung Hyun Choi, Xiaoxi Wang,Aaroshi Sahgal, Isaiah Udotong, Giulio Capolino, Tanya Bakshi, Allison Davanzo, Karen Pulido, andBailey Tregoning. The Appendix that accompanies the paper can be found on the author’s web page,together with the data and replication materials. The author declares that he has no relevant or materialfinancial interests that relate to the research described in this paper.

1See Cavallo (2013) and Cavallo and Rigobon (2016).2See Horrigan (2013), Griffioen, de Haan and Willenborg (2014), Boettcher (2015), Breton et al.

(2015), Krsinich (2015), Nygaard (2015), and Krsinich (2016).3See Chevalier and Goolsbee (2003), Brynjolfsson, Hu and Simester (2011), Edelman (2012), Cav-

allo, Neiman and Rigobon (2014), Gorodnichenko, Sheremirov and Talavera (2014), Simonovska (2015),Alvarez, Lippi and Le Bihan (2016), Cavallo (2016), and Gorodnichenko and Talavera (2016).

1

2 ARE ONLINE AND OFFLINE PRICES SIMILAR? - ALBERTO CAVALLO

place online. In fact, according to Euromonitor (2014), online purchases arecurrently less than 10 percent of all retail transactions in the US, and even lowerin other countries.

This paper provides the first large-scale comparison of online and offline pricesin large multi-channel retailers designed to answer this question. Using a combina-tion of crowdsourcing platforms, a mobile phone app, and web scraping methods,I simultaneously collected prices in both the online and offline stores of 56 ofthe largest retailers in 10 countries: Argentina, Australia, Brazil, Canada, China,Germany, Japan, South Africa, UK, and the United States. These data are usedto compare price levels, the behavior of price changes, and the selection of prod-ucts available for sale in the offline and online stores. I document country, sector,and retailer heterogeneity, and test whether online prices vary with ip-addresslocations or persistent browsing habits. The results have implications for NSOsand researchers using online data, as well as those interested in the effect of theInternet on retail prices.

The data collection effort is unprecedented in scope and size, and was carriedout as part of the Billion Prices Project (BPP). I first selected the retailers to besampled by focusing on the top 20 companies by market shares in each countrythat sell both online and offline (“multi-channel”), and have product barcodesthat can be matched across samples. Next, I used crowdsourcing platforms suchas Amazon Mechanical Turk, Elance, and UpWork to hire 323 workers to collectthe offline data. Each worker was assigned a simple task: to scan the barcodes andcollect prices for a random set of 10-50 products in any physical store of a givenretailer. In some cases they had to return to the same store multiple times to scanthe same set of products. Using a special app for android phones developed tosimplify and standardize the data collection process, these workers scanned eachproduct’s barcode, manually entered the price, took a photo of the price tag, andsent all the information via email to the BPP servers, where it was automaticallyprocessed and cleaned. A scraping software then used the barcode numbers tolook for the same product in the website of each retailer, and collected the onlineprice within a period of seven days. The matched online-offline dataset, availablefor download at bpp.mit.edu, contains prices for more than 24 thousand productsand 38 thousand observations sampled between December 2014 and March 2016.

The main finding is that online and offline price levels are identical about 72percent of the time, with significant heterogeneity at the country, sector, andretailer level. These percentages range from 42 percent in Brazil to 91 percentin Canada and the UK. The US is close to the average, with 69 percent. Atthe sector-level, drugstores and office-product retailers have the lowest share ofidentical prices, with 38 percent and 25 percent respectively, while in electronicsand clothing these numbers rise to 83 percent and 92 percent respectively. Whenthere is a price difference, the online markup tends to be small, with a magnitudeof -4 percent in the full sample. If I include observations with identical prices,the online price difference is only -1 percent on average.

ARE ONLINE AND OFFLINE PRICES SIMILAR? - ALBERTO CAVALLO 3

I also find that price changes have similar frequencies and sizes in the online andoffline data. However, only 19 percent of weekly price changes occur at the sametime. While this is higher than the unconditional probability of a simultaneousprice change, the individual price series are clearly not well synchronized.

The reasons for the existing online-offline price differences seem to vary acrossretailers and countries. Sales tend to create some discrepancies, with only 36percent of sale prices being identical across samples, but they have a small impactin the aggregate results because the number of sale observations is relatively small(11 percent of the total dataset). A similar thing happens with offline pricedispersion across physical stores, which tends to be low. Using a small sample ofoffline prices collected for multiple zip codes on the same day, I find that about78 percent of goods have the a single price within stores of the same retailer.I also found no evidence of “dynamic pricing” strategies that could potentiallycause online-offline differences. At least in the US, online prices do not changewith the location of the ip-address of the computer connecting to the website orwhen the scraping robot repeatedly browses the same webpage of a particulargood for a prolonged period of time. There is also no evidence that online-offlineprice differences are being driven by attempts to match the prices of Amazon.com,which are identical to the online prices in multi-channel retailers about 38 percentof the time.

In terms of product selection, 76 percent of the products sampled offline werealso found online by either using the automated scraping matching or by manuallysearching for the product description on the website. The price comparison resultsfor goods that can be automatically matched are similar to those that had to bemanually matched. There is also no evidence that retailers try to obfuscate theonline-offline price comparisons by changing the products’ identification numbers.

Despite the general similarity in online and offline prices, there is significantheterogeneity in pricing behaviors across retailers. Three main types of compa-nies stand out: those with nearly identical online and offline prices, those withstable online markups (either positive or negative), and those with different pricesthat are not consistently higher or lower online. Some of these patterns seem tobe sector-level behaviors, while others are common for most retailers within acountry.

For research economists using online data, these results provide evidence thatmost large multi-channel retailers price similarly online and offline. There areboth advantages and disadvantages of using online data, as I discuss in Cavallo(2016), but the ability to collect a massive amount of prices so cheaply providesunprecedented opportunities for economic research. My results suggest theseprices are valid sources of information for retail transactions, even those thattake place offline. Retailer heterogeneity, however, implies that researchers usingrelatively few sources of data should be cautious to understand particular pricingpatterns and control for any sampling biases.

For National Statistical Offices (NSOs), these results imply that the web can


be effectively used as an alternative data-collection technology to obtain the sameprices found offline. Prices collected through the web are very similar to thosethat can be obtained at a much higher cost by physically walking into a store.While many challenges to the use of online data in CPIs remain, such as themore limited sectoral coverage or the lack of quantity data, my results shouldhelp alleviate concerns about the peculiarities of prices collected online. TheBPP app and methodology developed in this paper are also publicly availableat bpp.mit.edu to be used for more country and retailer-specific validation tests,which are sensible given the high degree of heterogeneity in pricing behaviors.

Lastly, my findings have implications for people interested in the effects of theInternet on retail prices. The fact that online prices are the same for all locationsand also similar to offline prices collected from many different zip codes impliesthere is little within-retailer price dispersion. I also show this explicitly with someoffline data in multiple zip codes in Section 5.2. In practice, most retailers seemto have a single price for the majority of products, regardless of the location ofthe buyer and whether the product is sold online or at a particular offline store.This suggests that while the web has not reduced price dispersion across differentretailers, as documented by a large literature surveyed by Baye et al. (2006),it may have created incentives for firms to price identically in their own stores.This type of within-retailer price dispersion has received little attention in theliterature, even though it could have large welfare implications within countries.

This paper is related to a literature that studies the behavior of online prices.Some papers written in the early 2000s compared manually-collected prices of on-line retailers and traditional brick-and-mortar stores in a few narrow categoriesof goods. For example, Brynjolfsson and Smith (2000) compared prices for CDsand books in both online-only and multi-channel retailers (“hybrids” in their no-tation). They report that online prices were 9-16 percent lower and had smallerprice changes, but note that “findings would be strengthened if we excluded hy-brid retailers from our comparisons of price levels”, which implies that online andoffline prices for multi-channel retailers were closer together. Clay et al. (2002)also found similar prices for 107 books in both the websites and some physicalstores of Barnes & Noble and Borders, which is consistent with my results.4 Morerecent comparisons of online and offline prices expanded on the categories cov-ered but were limited to small ad-hoc samples in a few stores. Examples includeCavallo, Neiman and Rigobon (2014), Borraz et al. (2015), Cavallo, Neiman andRigobon (2015), and Cavallo (2016). A separate branch of the literature usesonline prices from “shopbots”, or price comparison websites, which are easier tocollect. Examples include Brynjolfsson and Smith (2001), Brynjolfsson, Dick andSmith (2009), Ellison and Ellison (2009a), Ellison and Ellison (2009b), Lunne-mann and Wintr (2011), Gorodnichenko, Sheremirov and Talavera (2014), andGorodnichenko and Talavera (2016). Although these papers do not directly com-

4For other papers in this literature, see Bailey (1998), Tang and Xing (2001), Clemons, Hann andHitt (2002), and Xing, Yang and Tang (2006).


pare prices with offline data, their results suggest that online prices change morefrequently and with smaller sizes than comparable findings in papers with offlineCPI prices. The difference with my findings is likely caused by their focus onretailers that participate in price-comparison websites. As Ellison and Ellison(2009a) discuss, such retailers face a uniquely competitive environment that cansignificantly affect their pricing behaviors.

I. Simultaneous Online-Offline Data Collection

A. Multi-Channel Retailers

There are many types of “online prices”, from those in marketplaces such asEbay, online-only retailers such as Amazon, and those sold by stores with bothan online and offline presence. In this paper, I focus on the prices of large “multi-channel” retailers that sell both online and offline. When considering all retailsales, this type of retailers still concentrate the vast majority of all retail transac-tions, making them the most important source of price data for applications thatrequire the use of “representative” data (such as inflation measurement). Despiteits importance, this is also the type of “online prices” that has received the leastattention in the academic literature due to lack of data. Furthermore, as pointedout by Brynjolfsson, Hu and Rahman (2013), technology is blurring the distinc-tions between physical and online retailing, making both traditional brick-and-mortar and online-only companies behave increasingly like multi-channel (“omni-channel”) retailers.

B. Retailer selection

The names of the retailers included in the data collection are shown in Table1. They satisfy three conditions. First, they are in the list of top 20 retailers bymarket share in their respective countries. The rank information was obtainedfrom Euromonitor International’s Passport Retailing Global Rankings. This helpsensure that I have a representative sample of the retail sector. Second, they sellboth online through a country-specific website and offline through physical stores.Most large retailers satisfy this condition. Third, there must be a way to perfectlymatch products online and offline. In practice, this means that the product idnumber collected offline can be used to find the product on the website.

C. Collecting Offline Prices in Physical Stores

Collecting prices offline is normally an expensive and complicated process.NSOs rely on a large number of trained data collectors to do it correctly. Un-fortunately, the micro data collected by NSOs for CPI purposes cannot be usedfor my comparisons because the retailer and product details are confidential in-formation. Lacking the budget for a traditional data collection effort, I looked


Table 1—Retailers Included

Country Retailers IncludedArgentina Carrefour, Coto, Easy, Sodimac, WalmartAustralia Coles, Masters, Target, WoolWorthsBrazil Droga Raia, Extra, Magazine Luiza, Pao de Azucar, RennerCanada Canadian Tire, Home Depot, The Source, Toys R Us, WalmartChina Auchan Drive, Sams ClubGermany Galeria Kaufhof, Obi, Real, Rewe, SaturnJapan Bic Camera, K’s Denki, Lawson, YamadaSouth Africa Clicks, Dis-Chem Pharmacy, Mr Price, Pick n Pay, WoolworthsUK Asda, Marks and Spencer, Sainsburys, TescoUSA Walmart, Target, Safeway, Stop&Shop, Best Buy, Home Depot,

Lowe’s, CVS, Macys, Banana Republic, Forever 21, GAP, Nike,Urban Outfitters, Old Navy, Staples, Office Max/Depot.

Notes: These retailers satisfy three conditions. First, they are in the list of top 20 retailers by mar-ket share in their respective countries according to Euromonitor International. Second, they sell bothonline through a country-specific website and offline through physical stores. Third, there is a way toperfectly match products online and offline for the price comparison. See the Appendix for more detailedcharacteristics and results.

for alternatives using new technologies. In particular, I rely on popular crowd-sourcing platforms, such as Amazon Mechanical Turk, Elance, and UpWork, tofind people willing to do simple data collection tasks. To minimize the chanceof data-entry errors, I developed a custom mobile phone app that simplified thedata collection process.

Crowd-sourcing platforms have many advantages. First, they allowed me tohire a large number of workers and reach multiple locations and cities within eachcountry. Second, with many workers I could limit the number of individual pricesthat each one of them had to collect. This reduced the burden on the worker andalso minimized the “show-rooming” concerns of the retailers. Showrooming is aterm used to describe the practice of visiting a physical store to examine a productbut later purchasing it online in another store. Some retailers worry about peoplewho use mobile apps to scan the product’s barcode and buy products online atother retailers, so if the data collectors spent too much time at each store, theymight be required to stop and asked to leave.5

Two main versions of the task were posted on the crowd-sourcing websites. Inthe simplest case, the worker had to use a mobile app provided by us to scan 10 to50 random offline products in any physical store, with some basic instructions tospread out the data collection across categories of goods. This provided the bulk

5I tried to conduct a similar large-scale offline data collection with MIT students in the Boston areain 2011, but most of them were asked to stop and leave the stores after a some time. Collecting datathis way appears to be easier now that more people use smartphones inside stores. Indeed, FitzGerald(2013) reports that fear of showrooming has faded for many US retailers. See Balakrishnan, Sundaresanand Zhang (2013) for an economic analysis of showrooming practices.


of the data that I use to compare price levels across samples. A more complexversion of the task required the worker to return to the same store every week fora full month and scan the same items. This gave me the panel of prices that Iuse to study price changes in Section III.

Figure 1. Screenshots from BPP App for Android Phones

The mobile app was custom-built to simplify and standardize the data collectionprocess. It is an app for android phones called “BPP @ MIT”, available fordownload at Google’s Play Store.6 Every time a worker visits a store, she clickson a button to open a new file. For the first product, she has to enter the store’sname, zip code, and country. Then she scans the UPC barcode of the product(or the barcode on the price tag, depending on the particular retailer instructionsprovided), manually enters the price shown in the price tag next to the product(including all sales displayed), marks the price as “regular” or “sale”, and takesa photograph of the price tag (which can be used to detect errors and validatethe data). All products are scanned in a loop which makes the process quick andsimple. When done, the worker presses another button to email the data to theBPP servers. A member of the BPP team verifies the submitted data and paysthe worker.

Every few hours, the BPP servers automatically processed the incoming offlinefiles to clean and consolidate the data for each retailer. The offline barcodeinformation is then used to collect the online price in the retailer’s website, asdescribed below.

6See https://play.google.com/store/apps/details?id=com.mit.bpp. The app can be downloaded forfree but a “project code” must be requested to the BPP team. This code is used to separate the datafrom different projects. See http://bpp.mit.edu/offline-data-collection/ for more details.


D. Collecting Online Prices on each Retailer’s Website

To collect online prices, I built a custom scraping “robot” for each retailer.These robots are specialized software that are programmed to use the productbarcode to query the retailer’s website and collect the online price and otherproduct information. In most cases, the robot was designed to use the website’ssearch box to enter the product id obtained offline. For more general details onthe BPP’s online scraping methods, see Cavallo and Rigobon (2016).

The price collected online is the posted price for the product on the retailer’swebsite, including any sales or discounts that apply to all customers. Whethertaxes are added or not depends on the display conventions for prices in eachcountry, but the same condition applies both online and offline. For example, USprices include sales but are typically shown without taxes, both on the websiteand the price tags found in physical stores. In all other countries, sales or VAT taxrates are usually included in the price in both locations. Shipping costs are neverincluded in these online prices, so my comparisons are for posted prices excludingshipping costs. Retailers have different ways to charge for shipping. The mostcommon is a set of shipping fees that varies with the total amount of the sale orweight of the products. Some retailers offer free shipping, which could mean thatthey adjust their online prices to compensate. The results at the retailer levelprovide information that can be used to determine when this is happening.

Nearly all of the online retailers in the sample have a single price online foreach product, independent of the location of the buyer. In other words, someonepurchasing a laptop from Bestbuy in San Francisco sees the same price as someonedoing it from Boston. The only exceptions are supermarkets, which sometimesrequire buyers to enter their zip code or location before displaying prices. Thereare only five retailers that do this in my sample. I always use the same zipcode when collecting data online, independently of the one where the offline pricewas obtained, so this can cause some price level differences between the onlineand offline data for those retailers. However, in the Appendix I use a scrapingexperiment with one of the largest US supermarkets to show that even retailersthat ask for zip code information tend to price their goods identically in mostlocation. Furthermore, removing this type of retailers has little impact on myaggregate results.

For all benchmark results, I allow online prices to be collected within 7 daysof the offline price and also exclude sale prices. Results are similar for pricescollected on the same day, or including sale prices, as shown in the Appendix.

E. The Online-Offline Matched Data

Table 2 shows the main characteristics of the matched data. I collected prices in56 retailers for more than a year, between December 2014 and March 2016. Thereare more than 24 thousand products and 38 thousand observations in total. Thisdataset can be downloaded from http://bpp.mit.edu, together with the replication


scripts for the results below.

Table 2—Data by Country

(1) (2) (3) (4) (5) (6) (7)Country Retailers Start End Workers Zip Codes Products ObservationsArgentina 5 02/15 08/15 18 23 2324 3699Australia 4 03/15 08/15 13 22 3073 3797Brazil 5 05/15 03/16 18 26 1437 1915Canada 5 12/14 07/15 15 45 2658 4031China 2 07/15 03/16 5 6 410 513Germany 5 03/15 03/16 9 20 1215 1604Japan 4 04/15 03/16 7 23 1127 2186South Africa 5 03/15 03/16 21 31 2336 3212UK 4 03/15 05/15 12 32 1661 2094USA 17 12/14 03/16 206 274 7898 15332All Countries 56 12/14 03/16 323 499 24132 38383

Notes: Column 1 has the number of retailers. Columns 2 and 3 have the start and end months of datacollection. Columns 4 and 5 report the number of workers that collected the data and zip codes withoffline prices. Columns 6 and 7 provide the number of products and price observations that could bematched with both online and offline information. Details by retailer are provided in the Appendix.

The data coverage varies across countries. The effort was concentrated in theUS, with 17 retailers and about forty percent of all observations. On the otherextreme is China, with only two retailers. I was unable to expand the offline datacollection in China because large retailers explicitly prohibit taking photographsand recording prices at physical locations. Apparently, “showrooming” is moreextended in China, and therefore retailers try to prevent the use of mobile phonesin their stores. A survey conducted by IBM in 2013 found that about 24 percentof people in China admitted to having visited a physical store to buy online,compared with only 4 percent in the United States.7

II. Price Levels

Table 3 compares the price levels across the online and offline samples. Column3 shows the percentage of observations that have identical online and offline pricesup to the second decimal.

The percentage of identical prices is 72 percent for all pooled observationsand also for the average across countries. Some countries, such as Japan, havepercentages close to 50 percent, while other such as Canada and the UK haveover 90 percent of all prices being identical online and offline. The US is close tothe average, with 69 percent of identical prices.

Columns 4 and 5 show the share of prices that are either higher or lower online.Conditional on a price difference, most countries tend to have lower online prices,

7Klena and Puleri (2013).


Table 3—Country - Price Level Differences

(1) (2) (3) (4) (5) (6) (7)Country Retailers Observations Identical Higher Lower Online Online

(percent) Online Online Markup Difference(percent) (percent) (percent) (percent)

Argentina 5 3699 60 27 13 3 1Australia 4 3797 74 20 5 5 1Brazil 5 1915 42 18 40 -7 -4Canada 5 4031 91 3 5 -5 0China 2 513 87 7 6 3 0Germany 5 1604 74 4 23 -8 -2Japan 4 2186 48 7 45 -13 -7South Africa 5 3212 85 6 9 -3 -1UK 4 2094 91 2 7 -8 -1USA 17 15332 69 8 22 -5 -1All Countries 56 38383 72 11 18 -4 -1

Notes: Column 3 shows the percentage of observations that have identical online and offline prices.Column 4 has the percent of observation where prices are higher online and column 5 the percentageof price that are lower online. Column 6, is the online markup, defined as the average price differenceexcluding cases that are identical. Column 7 is the average price difference including identical prices.

with the exception of Argentina and Australia. The three countries with thelowest percentages of identical prices, where differences matter the most, tendto have heterogeneous behaviors. In Argentina, non-identical prices tend to behigher online, with an average markup of 3 percent. In Brazil, they are lower,with a markup of -7 percent. Japan is an outlier, with prices that are lower online45 percent of the time, with an average markup of -13 percent.

The average size of the price differences is quite small. This can be seen inColumns 6 and 7, where a positive number means that prices are higher online.Column 6 shows the online “Markup”, excluding cases where prices are identical,while column 7 shows the online “Difference”, which includes cases with no pricedifference. The online markup tends to be small, with a magnitude of -4 percentin the full sample. Adding prices that are identical makes the online-offline pricedifference only -1 percent on average.

Overall, these results show little difference between online prices collected fromthe website of multi-channel retailers and the offline prices that can be obtainedby visiting one of their physical stores.

The aggregate results, however, hide important heterogeneity at the sector level.Table 4 shows similar results for retailers grouped by the type of good they sell.

Drugstores and office-supply retailers have the lowest share of identical pricesonline and offline. For office products, prices are sometimes higher and some-times lower online, without any clear patterns, as if the stores were managedindependently. Drugstores, by contrast, tend to have lower prices online, possiblybecause they are “convenience” stores such as CVS and Walgreens in the US thatcan charge higher prices to offline customers.


Table 4—Sector - Price Level Differences

(1) (2) (3) (4) (5) (6) (7)Sector Retailers Observations Identical Higher Lower Online Online


Food 10 5953 52 32 15 3 1Clothing 7 2534 92 5 3 3 0Household 9 7875 79 5 16 -8 -2Drugstore 4 3053 38 11 52 -5 -3Electronics 5 3712 83 4 13 -9 -1Office 2 1089 25 37 38 1 1Multiple/Mix 18 14149 80 5 15 -9 -2

Notes: Data classified into sectors at the retailer level. Column 3 shows the percentage of observationsthat have identical online and offline prices. Column 4 has the percent of observation where prices arehigher online and column 5 the percentage of price that are lower online. Column 6, is the online markup,defined as the average price difference excluding cases that are identical. Column 7 is the average pricedifference including identical prices.

Electronics and clothing have the highest share of identical prices. For cloth-ing, prices are basically the same, with most of the observed differences possiblycoming from offline data collection errors. For electronics, prices are lower online13 percent of the time, with an average markup of -9 percent (the highest in thissample).

Figure 2 shows the histograms for non-zero price differences in each country.The cases of Argentina and Australia stand out because there are spikes aroundthe 5 percent magnitude of differences. This is caused by stable markups in onlineprices for some of the largest retailers. In all other countries, the price differencesare more dispersed in the range of -50 percent to 50 percent.

As pointed out by Nakamura and Steinsson (2008), sale events are frequent insome countries, and the magnitude of the price changes that they generate can belarge. I do find that sale prices create more differences between online and offlinesamples, the share of identical online and offline prices for sale observations beingonly 36 percent. But this has little impact on the full-sample results becausethe number of sales is small: only 11 percent of all matched observations haveeither an online sale (4.12 percent), an offline sale (5.03 percent), or both (1.92percent).8

Similarly, restricting the sample to include only prices collected on the sameday (instead of allowing for a 7-day window) has little impact on the main results.The reason is that prices do not typically change more than once a week. Detailsare provided in the Appendix.

Another potential reason for some of the price level differences is that goods

8My ability to control for sales is somewhat limited because workers could not identify offline saleswith the app until October 2015, and some of the scrape jobs were not able to include online saleindicators. It is therefore possible that the main results still contain a lot of sales that I cannot controlfor, and the share of identical prices would rise significantly if these observations were removed.


Figure 2. Histograms of Non-Zero Price Level Differences

010

2030

4050

Perc

ent

-50 0 50Difference (%)

ARGENTINA

010

2030

4050

Perc

ent


AUSTRALIA

010

2030

4050

Perc

ent


BRAZIL

010

2030

4050

Perc

ent


CANADA

010

2030

4050

Perc

ent


CHINA0

1020

3040

50Pe

rcen

t


GERMANY

010

2030

4050

Perc

ent


JAPAN0

1020

3040

50Pe

rcen

t


SOUTHAFRICA

010

2030

4050

Perc

ent


UK

010

2030

4050

Perc

ent


USA

Notes: Price differences excluding identical prices. A positive number means that the online price ishigher than the offline price. Histogram scales are matched across countries. Bin width is 1 percent.

have prices with similar time series but are not synchronized. I look for directevidence of this in the next section, by comparing online and offline changes fora smaller sample of goods for which I have multiple weekly observations.

III. Price Changes

This section compares the behavior of price changes in the online and offlinesamples. A price change is computed here as a non-zero log difference in theprice between weeks t and t+1. I study the frequency, size, and timing of pricechanges.

Table 5 shows that the frequency of online and offline prices changes is quitesimilar. The first two columns show the number of observations and price changes.There are fewer observations than in previous sections because I have a shorttime series for a limited subset of goods, and only about 10 percent of thoseobservations have a price change. The frequency statistics reported in columns 3and 4 are computed for each individual good first (as the share of observationswith a price change), and then averaged across countries. Column 5 shows the


Table 5—Country - Price Change Frequency and Size

(1) (2) (3) (4) (5) (6) (7) (8)Observations Price Mean Mean Equality Mean Mean Equality

Changes Frequency Frequency Test Absolute Absolute TestOnline Offline p-value Size Size p-value

Online Offline(percent) (percent)

Argentina 1392 245 0.137 0.146 0.56 13.61 12.46 0.57Australia 759 72 0.056 0.090 0.07 45.76 42.62 0.67Brazil 483 85 0.167 0.138 0.36 10.55 9.36 0.53Canada 1427 120 0.077 0.068 0.48 31.11 21.71 0.06Germany 419 16 0.035 0.041 0.74 27.08 15.86 0.26Japan 1071 98 0.074 0.014 0.00 12.10 8.20 0.34South Africa 882 109 0.100 0.077 0.17 23.33 16.99 0.11UK 429 25 0.046 0.070 0.28 47.68 41.78 0.67USA 7505 563 0.052 0.046 0.33 23.78 21.31 0.20All Countries 14367 1328 0.076 0.068 0.07 22.02 19.94 0.10

Notes: China is excluded due to lack of price change data. The first two columns show the number ofobservations and price changes. The frequency statistics reported in columns 3 and 4 are computed foreach individual good as the share of observations with a price change, and then averaged across countries.Column 5 shows the p-value of a two-sided t-test with a null hypothesis of equal average frequencies inthe online and offline samples. Columns 6 and 7 report the mean absolute size of price changes. Column8 is the p-value of a two-sided t-test of equality in the online and offline means.

p-value of a two-sided t-test with a null hypothesis of equal average frequenciesin the online and offline samples. I can only reject the null of equality with someconfidence in the cases of Australia and Japan. Although the full sample resultsappear to have slightly more frequent changes online, this is entirely driven bythe data from Japan.

In addition to similar frequencies, online and offline price changes tend to havesimilar sizes. This can be seen in columns 6 and 7, where I report the meanabsolute size of price changes. Column 8 is again the p-value of a two-sided t-test of equality in the online and offline means. The null hypothesis can only berejected in Canada, where online price changes seem to be larger. In all othercountries, the difference is not statistically significant.

Similar frequencies and sizes do not imply that price changes are perfectlysynchronized. This can be seen in Table 6, which focuses on the timing of changes.Price changes can occur online, offline, or in both locations. Column 3 reports thepercentage of price changes for a given product that occur both online and offlineat the same time, which I refer to as “synchronized”. Only 19 percent of the 1328price changes were synchronized across online and offline samples. While this ishigher than the unconditional probability of a simultaneous price change shownin column 4 (using the unconditional frequencies and assuming independence),these price series are still far from being perfectly synchronized.

Overall, these results suggest that the online and offline price series behavesimilarly but are not perfectly synchronized. In a related paper, Cavallo andRigobon (2016), we find evidence that online price inflation tends to anticipateoffline CPI inflation. A faster adjustment to shocks could be the reason why


Table 6—Synchronized Price Changes

(1) (2) (3) (4)Observations Price Changes Synchronized Unconditional

Price Changes Probability(percent) (percent)

Argentina 1392 245 35 2.0Australia 759 72 22 0.5Brazil 483 85 18 2.3Canada 1427 120 32 0.5Germany 419 16 31 0.1Japan 1071 98 1 0.1South Africa 882 109 15 0.8UK 429 25 44 0.3USA 7505 563 11 0.2All Countries 14367 1328 19 0.5

Notes: China is excluded due to lack of price change data. Column 3 reports the percentage of pricechanges for a given product that occur both online and offline at the same time, which I refer to as“synchronized”. The unconditional probability of a synchronized price change in column 4 is obtainedby multiplying the frequencies of price change in Table 5.

online price changes are not synchronized with offline changes. Unfortunately,the limited panel data available so far does not allow me to explicitly test thishypothesis in this paper.

IV. Other Reasons for Online-Offline Differences

In this section I consider three other potential reasons for the differences be-tween online and offline prices that required a special data-collection effort: differ-ent online prices based on ip-address or persistent browsing habits, multiple offlineprices in different physical stores, and attempts to match prices at Amazon.com

A. IP-Address Location and Persistent Browsing

There have been reports suggesting that some retailers change online pricesbased on the browsing habits of the consumer or the location associated withthe ip address of the computer being used to purchase online. See, for example,Valentino-DeVries, Singer-Vine and Soltani (2012), Mikians et al. (2012), andMikians et al. (2013). If these pricing behaviors are common for the multi-channelretailers in my sample, they could help explain some of the price level differencesin the data. To test whether prices vary with browsing habits or ip address, I rantwo experiments with special versions of the scrape robots for US retailers.

The first experiment was designed to test whether prices change based on thezip code associated with the ip address of the computer collecting the data. IP


addresses are unique numeric identifiers for computers that are connected to anetwork. They are assigned by internet service providers and have an associatedgeographical location that is public information. For example, MIT’s campus ipaddresses range from 18.0.0.0 to 18.255.255.255 and are geographically linked tothe 02139 zip code in Cambridge, Massachusetts. In principle, retailers coulddetect the ip address of the consumer visiting a site and automatically change theprices displayed based on its geo-location information. To test if this is happening,I randomly selected 5 products in each of the 10 US retailers and scraped theirprices twelve times in a consecutive loop. In each loop, I changed the ip addressof the robots by using 12 different proxy servers in 9 US cities (Atlanta, Burbank,Charlotte, Chicago, Cleveland, Miami, Nashville, New York, and two proxies inPhoenix) and 2 international locations (Canada and UK).9 I did not find anyevidence of this type of price discrimination. In all cases, prices were the same fora given product, regardless of what ip address was used to connect to the retailerwebsites.

The second experiment was designed to test if frequent visits to the webpage ofa particular product could lead the retailer to change the price displayed. In thiscase, I scraped a single product in each retailer every five minutes for a full day.Once again, there was no evidence of price discrimination based on persistent-browsing habits: prices were always the same.

While these forms of online price discrimination may be important in otherindustries (for example airlines and hotels), my results suggest that they are notcommonly used in large multi-channel retailers in the US. A likely reason is thatretailers may fear antagonizing their customers if reports of these tactics were tobecome publicized in the press, as it famously happened in 2000 with Amazon’spricing tests.10

B. Offline Price Dispersion

Most retailers have a single price online regardless of the location of the buyer,so a second potential reason for online-offline differences may be that there aresome prices differences across physical stores.

To test for the effects of offline price dispersion, I use a small subset of productsfor which I have offline prices for multiple zip codes collected on the same day.These data include 406 observations in 9 retailers and 46 zip codes in the US.Table 7 shows the results for the online - offline comparison restricted to thismulti-zipcode dataset.

9A proxy server is a computer that acts as an intermediary for the communications between twoother computers in a network –in this case between the machine where the scraping software runs andthe server hosting the website of the retailer–. From the retailer’s website perspective, the request wascoming from the ip address associated with the proxy server.

10See CNN (2000) and Valentino-DeVries, Singer-Vine and Soltani (2012) for a more recent example.A pricing strategy that appears to be more common than price discrimination is called “steering”, wherethe retailer changes the order or ranking of goods shown to customers based on their location or browsingcharacteristics. See, for example, Mattioli (2012).


Table 7—Online - Offline Price Level Differences for Multiple Zipcodes

(1) (2) (3) (4) (5) (6) (7)Country Retailers Observations Identical Higher Lower Online Online


USA 9 406 60 11 29 -4 -2Different Offline 7 85 35 16 48 -5 -3Identical Offline 8 316 67 9 24 -3 -1

Notes: Column 3 shows the percentage of observations that have identical online and offline prices.Column 4 has the percent of observation where prices are higher online and column 5 the percentageof price that are lower online. Column 6, is the online markup, defined as the average price differenceexcluding cases that are identical. Column 7 is the average price difference including identical prices.

There are several things to note here. First, even though the sample is small,we get roughly the same share of identical online-offline prices that in Table 3 ofthis paper, with 60 percent of the prices being identical online and offline. Second,as expected, goods that have different offline prices across zip codes tend to havemuch lower probability of identical online - offline prices, about 35 percent ofthe time. Third, if we focus exclusively on products with the same offline priceeverywhere, labeled ”Identical Offline”, the percentage of identical online - offlineprices rises from 60 percent to 67 percent.

While offline price dispersion can create online-offline price differences, the im-pact is limited because there is not much offline dispersion to begin with. Indeed,about 78 percent of products sampled have the same price in different physicalstores within the same retailer, as seen in Column 2. Sector results range from66 percent in drugstores to 96 percent in electronics, consistent with the sectoraldifferences in the online-offline comparison in Section II.11 In the Appendix Ifurther show that a large multi-channel supermarket that explicitly asks onlinebuyers to enter their zip codes also tends to limit the amount of price dispersionacross locations. Overall, these results reinforce the finding that price dispersionis low for both online and offline prices within multi-channel retailers.

To some readers, the lack of offline price dispersion may appear to be at oddswith a growing literature that uses scanner data and documents a significant pricedifferences across physical stores. For a recent example, see Kaplan and Menzio(2015). There are many reasons that can explain the apparent differences withmy results. First, many papers in this literature compare data from differentretailers, so that within retailer price dispersion is mixed with between retailerprice dispersion. My results focus exclusively on price differences within retailers.Second, the price in scanner datasets is typically a weekly average. As I discuss inCavallo (2016), this can cause significant measurement error for some applications.For example, consider a good with identical prices in two stores, a price change

11See the Appendix for more details as well as results from a larger dataset that includes offlineobservations for which no online price is available.


on a Wednesday, and a single transaction in each store. If one store sold thegood on a Monday, and the other on Friday, the “weekly” price will appear tobe different when in fact prices were identical on a daily basis. Similarly, somescanner datasets tend to have unit values instead of prices. These are calculatedas the ratio of sales to quantities sold, and can therefore be affected by the numberof coupons used or the share of transactions that take place at different prices.Of course, for some purposes it makes sense to include coupons or transactionweights that affect the price actually payed by the consumer, but the fact thatthere is price dispersion caused by coupons should not lead us to believe thatprices for the same goods are shown with different prices across stores of the sameretailer. Third, price dispersion is often measured within a month or a quarter,so much of difference in observed prices is caused by the same good being boughtat different times. Finally, most scanner datasets contain prices for groceries andrelated goods. These are also the sectors for which I find more online-offline pricedispersion, as well as offline price differences across physical stores.

C. Amazon Pricing

A third potential reason for differences in online and offline prices is that multi-channel retailers may be matching their online prices to those in online-only re-tailers such as Amazon.com, and by doing so, they create a wedge with the pricesat their physical stores.

To test this possibility, I created a special dataset that contains three pricesfor each product: the offline price at a multi-channel retailer, the online price inthe same retailer, and the price at Amazon.com. The matched data contain 1361observations from 455 products and 8 multi-channel retailers: BestBuy, Walmart,Target, Lowes, Macys, OfficeMax, and Staples. Amazon’s prices considered belowcorrespond to those products marked as ”Sold by Amazon.com”. To be consistentwith the rest of the paper, I focus on prices collected within up to seven days andexcluding sales. More details on how this data was collected, as well as results forproducts with sales or sold by third-party sellers are provided in the Appendix.

Figure 3 compares Amazon’s prices separately to both the offline and onlineprices from multi-channel firms. A large share of prices are identical in both cases,which is surprising given that this is comparing prices across different retailers.As expected, Amazon’s prices are closer to the online prices. They are identicalto the online prices approximately 38 percent of the time, and the average pricedifference is -5 percent. The same estimates for the Amazon - offline comparisonare 31 percent and -6 percent respectively.

This does not mean that multi-channel retailers are making their online andoffline prices different in order to match the online price to Amazon’s one. In fact,as Table 8 shows, the conditional probability of having an identical online pricewith Amazon is roughly the same for goods with identical online-offline prices thanfor those that have some online-offline price difference. The same conclusion canbe obtained by running a simple probit regression of an identical online-offline


Figure 3. Price Differences with Amazon.com (US only)

010

2030

40Pe

rcen

t

-100 -50 0 50 100Price Diference (%)

Amazon vs Offline Amazon vs Online

Notes: Price difference in Amazon.com prices relative to the offline and online prices from multi-channelretailers obtained from 1361 observations from 455 products and 8 multi-channel retailers: BestBuy,Walmart, Target, Lowes, Macys, OfficeMax, and Staples. A negative number means Amazon is cheaper.Results for products marked as ”Sold by Amazon.com” are shown here, with prices collected within upto seven days and excluding sales. More details and results for products with sales or sold by third-partysellers at Amazon’s ”Marketplace” are provided in the Appendix.

price on an identical amazon-online price. There is no statistically significantrelation between these two variables. The only indication that Amazon’s pricesmatter for the online-offline price differences is found in columns 8 and 9, whichshow that the difference with Amazon’s prices is smaller for goods that are notidentical within the multi-channel retailers.

V. Product Selection

The similarity between online and offline prices in previous sections would havedifferent implications if most goods sold offline were not available online. I there-fore now estimate the “overlap” in product selection across samples, defined asthe share of offline goods that are also available online.12

12Note that, given the data characteristics, I can only estimate how many offline products are alsosold online, but not the other way around. In some retailers, the online selection of goods can be larger


Table 8—Amazon - Online Price Level Differences

(1) (2) (3) (4) (5) (6) (7)Retailers Observations Identical Higher Lower Amazon Amazon

(percent) Amazon Amazon Markup Difference(percent) (percent) (percent) (percent)

All Observations 8 1049 38 14 47 -9 -5Identical On-Off 8 801 38 11 51 -10 -6Different On-Off 8 248 38 25 37 -3 -2

Notes: There are 312 observations with an Amazon price and either an online or offline price, but notboth, so they are excluded from these results. Column 3 shows the percentage of observations that haveidentical Amazon and online prices at multi-channel retailers. Column 4 has the percentage of pricesthat are higher in Amazon and column 5 the percentage of prices that are lower in Amazon relative tothe online prices. Column 6, is the Amazon markup, defined as the average price difference excludingcases that are identical. Column 7 is the average price difference including identical prices.

In principle, I could use the 63 percent of offline barcodes received through theapp for which the scraping software found data online. The problem with thisnumber, however, is that the automated matching procedure can fail for manyreason: the worker may scan the wrong barcode, the app can incorrectly readthe barcode, or the scraping robot can fail while checking the website. To geta better estimate of the overlap degree, we manually checked how many of theoffline products could also be found online for a sample of 100-200 observationsper retailer using all the information submitted by the workers, including theproduct description readable in the photo of the price tag. The results, groupedby country, are reported in Table 9.

As can expected given the large product variety in these websites, a large frac-tion of goods found offline can also be found available online. On average, 76percent of all products randomly collected at the physical stores could also befound on the retailer’s website. There are important differences among countries,although they seem to be unrelated to the findings in previous sections. Chinaand Germany have the lowest overlap, while Australia, Brazil, and the UK thehighest. In the US, 81 percent of offline products were also found online.

Furthermore, both the automatic and manually-matched goods produced simi-lar results for online and offline price-level comparisons, as shown in the Appendix.This rules out the possibility that goods that could not be automatically matchedwere precisely those where the online and offline prices are different. This wouldhappen, for example, if retailers changed the online id number for those goodsas a way to obfuscate their price differences and prevent any comparisons. Theevidence suggests that this is not generally the case.

than in a single physical store because online sales can be shipped from large centralized warehouses.See Quan and R. (2014) for a recent discussion of the welfare effect of online and offline product variety.


Table 9—Retailer - Product Selection Overlap

(1) (2) (3) (4)Country Sample Found Found Total

Automatically Manually Overlap(percent)

Argentina 500 294 52 73Australia 500 435 36 95Brazil 400 331 12 86Canada 500 279 132 85China 100 50 3 53Germany 400 178 23 52Japan 500 329 61 74South Africa 500 332 60 76UK 500 373 59 86USA 1600 1003 316 81All Countries 5500 3604 754 76

Notes: We took a random sample of 100-200 offline prices per retailer and manually searched for thesame products in the corresponding website. Column 2 shows the number of products that were foundonline by the automated process used to build the matched dataset in the paper. Column 3 shows thenumber of products that were missed by the automated process but were found online by manuallychecking the website. Column 4 shows the estimate for the total overlap in product selection. Only asubset of retailers in each country are included. These numbers are lower-bound estimates for the overlapbecause some of our manual checks took place several days after the original offline data was collected.Results by retailer are provided in the Appendix.

VI. Retailer Heterogeneity

The country-level results in the previous sections conceal a great deal of het-erogeneity across retailers in each country. Details for each retailer can be seenin Appendix Table A1, were I show price level and changes results for all retailerswith at least 100 observations.

Three main types of retailers are typical. First, there are retailers where onlineand offline prices are identical most of the time. These are cases where theretailer explicitly chooses to have the same online and offline price. Second, thereare also some retailers with a low share of identical prices, but no clear onlinemarkups. Many retailers in Brazil, for example, exhibit this pattern. These arelikely cases where the online store is simply treated as another outlet, sometimescheaper, sometimes more expensive. Third, there are retailers with a low shareof identical prices and a significant online markup (either positive or negative).There are some examples in Argentina, Brazil, Japan, and the US. These patternsmay reflect a desire to compensate for shipping costs or price-discriminate onlineconsumers.

Whether each kind of retailer is useful as a source of data depends on thepurpose of the paper or application. For example, using online prices for the


retailer in Argentina where 79 percent of prices are higher online is not a problemfor measuring inflation as long as the online markup is relatively constant overtime, but it would bias the results if we were interested in comparing price leveldifferences across countries. Unless a correction is applied, the online data wouldmake prices in Argentina appear higher than what they really are. Identifyingthese special patterns and correcting for any biases is particularly important inpapers or applications that use online data from one (or a few) retailers.

VII. Conclusions

This paper shows that in large multi-channel retailers there is little differencebetween the online price collected from a website and the offline price obtained byvisiting the physical store. Prices are identical about 72 percent of the time, andwhile price changes are not synchronized, they have similar frequencies and sizes.At the same time, there is considerable heterogeneity across countries, sectors,and retailers.

For research economists using online data for macro and international researchquestions, my results provide evidence that online prices are a representativesource of retail prices, even if most transactions still take place offline. At amore micro level, the differences in behaviors can be used to better model thepricing dynamics and strategies of different types of retailers in various sectorsand countries. This high degree of heterogeneity also implies that papers that userelatively few sources of data should be cautious to understand relevant pricingpatterns and control for any potential sampling biases.

For National Statistical Offices (NSOs) considering the use of online data forconsumer price indexes, my results show that the web can be effectively used as analternative data-collection technology for multi-channel retailers. Particularly forproducts such as electronics or clothing, the price collected on the web will tendto be identical to the one that can be obtained by walking into a physical store.Online prices are not only cheaper to collect, but they also provide informationfor all goods sold by each retailer, with many details per product, uncensoredprice spells, and can be collected on a high-frequency basis without any delays. Ofcourse, there are also many potential disadvantages of using online data, includinglimited sector coverage and the lack of information on quantities, as we discuss inCavallo and Rigobon (2016). But from a data-collection perspective, my resultssuggest that the online-offline price differences should not be a major source ofconcern.

For those interested in the effect of the Internet on retail prices, my findingsimply little within-retailer price dispersion, both online and offline. While theInternet may not have reduced dispersion across retailers, it seems to have cre-ated the incentives for companies to price identically across their own physicaland online stores. More research is needed to understand the mechanisms thatdrive this effect. One possibility is that retailers are worried about antagonizingcustomers who can now easily compare prices online through the web or their


mobile phones. This might even be affecting cross-country pricing, as suggestedby Cavallo, Neiman and Rigobon (2014), where we found evidence that globalfirms such as Apple and Ikea tend to price identically in countries that use thesame currency, where it is trivial for consumers to compare prices across borders.

Future work should also try to understand why there are still some observedprice-level differences. One explanation may be that online prices adjust faster toshocks. That would be consistent with the un-synchronized price change resultsin this paper and the anticipation in online price indices documented in Cavalloand Rigobon (2016). Another potential reason is that location-specific sales oroffline price dispersion may play a larger role than I can detect in these data. Inparticular, the offline price comparisons for multiple zip codes in Section IV.Bcould be expanded to cover more sectors and countries. In addition, good-levelcharacteristics, such as the bargaining power of the manufacturer or the natureof its production and distribution costs, may help explain why some goods haveidentical prices while others do not.

Another limitation of my analysis is the lack of quantity information at theproduct and retailer levels. For some applications, such as the computationsof price indices, we can use category weights in official CPI data. But otherpricing statistics may change considerably when individual goods are weightedby sales, as shown with online book sales by Chevalier and Goolsbee (2003).Future research should try to combine online prices with other micro data, suchas scanner datasets, that can provide more detailed quantity information.

Finally, with the exception of the Amazon results in Section IV.C, this paperdoes not study the prices of online-only retailers or small companies that par-ticipate in online “marketplaces”. If their share of retail transactions continuesto grow, a large-scale comparison with traditional multi-channel retailers will beneeded to better understand how pricing strategies and dynamics are likely toevolve in the future.

REFERENCES

Alvarez, Fernando, Francesco Lippi, and Herve Le Bihan. 2016. “Thereal effects of monetary shocks in sticky price models: a sufficient statisticapproach.” American Economic Review, Forthcoming.

Bailey, J. 1998. “Electronic Commerce: Prices and Consumer Issues for ThreeProducts: Books, Compact Discs and Software.” OECD Publishing OECDDigital Economy Paper 32.

Balakrishnan, Anantaram, Shankar Sundaresan, and Bo Zhang. 2013.“Browse-and-Switch: Retail-Online Competition under Value Uncertainty.”Production and Operations Management Society, 23(7): 1129–1145.

Baye, Michael R., John Morgan, Patrick Scholten, and others. 2006.“Information, search, and price dispersion.” Handbook on economics and infor-mation systems, 1: 323–375.


Boettcher, Ingolf. 2015. “Automatic data collection on the Internet.” StatisticsAustria.

Borraz, Fernando, Alberto Cavallo, Roberto Rigobon, and LeandroZipitria. 2015. “Distance and Political Boundaries: Estimating Border Effectsunder Inequality Constraints.” International Journal of Finance & Economics,21(1): 3–35.

Breton, Robert, Gareth Clews, Liz Metcalfe, Natasha Milliken,Christopher Payne, Joe Winton, and Ainslie Woods. 2015. “Researchindices using web scraped data.” Office for National Statistics UK.

Brynjolfsson, Erik, and Michael D. Smith. 2000. “Frictionless Commerce?A Comparison of Internet and Conventional Retailers.” Management Science,46(4): 563–585.

Brynjolfsson, Erik, and Michael D. Smith. 2001. “The Great Equalizer?Consumer Choice Behavior at Internet Shopbots.” Social Science Research Net-work SSRN Scholarly Paper ID 290323, Rochester, NY.

Brynjolfsson, Erik, Astrid Andrea Dick, and Michael D. Smith. 2009.“A Nearly Perfect Market?” Social Science Research Network SSRN ScholarlyPaper ID 450220, Rochester, NY.

Brynjolfsson, Erik, Yu Hu, and Duncan Simester. 2011. “Goodbye paretoprinciple, hello long tail: The effect of search costs on the concentration ofproduct sales.” Management Science, 57(8): 1373–1386.

Brynjolfsson, Erik, Yu Jeffrey Hu, and Mohammad S Rahman. 2013.“Competing in the Age of Omnichannel Retailing.” MIT Sloan ManagementReview, 54(4): 23–29.

Cavallo, Alberto. 2013. “Online and official price indexes: Measuring Ar-gentina’s inflation.” Journal of Monetary Economics, 60(2): 152–165.

Cavallo, Alberto. 2016. “Scraped Data and Sticky Prices.” Review of Economicsand Statistics, Forthcoming.

Cavallo, Alberto, and Roberto Rigobon. 2016. “The Billion Prices Project:Using Online Data for Measurement and Research.” Journal of Economic Per-spectives, 30(2): 151–78.

Cavallo, Alberto, Brent Neiman, and Roberto Rigobon. 2014. “CurrencyUnions, Product Introductions, and the Real Exchange Rate.” Quarterly Jour-nal of Economics, 129(2).

Cavallo, Alberto, Brent Neiman, and Roberto Rigobon. 2015. “The PriceImpact of Joining a Currency Union: Evidence from Latvia.” IMF EconomicReview, 63(2): 281–297.


Chevalier, Judith, and Austan Goolsbee. 2003. “Measuring prices and pricecompetition online: Amazon. com and BarnesandNoble. com.” Quantitativemarketing and Economics, 1(2): 203–222.

Clay, Karen, Ramayya Krishnan, Eric Wolff, and Danny Fernandes.2002. “Retail strategies on the web: Price and nonprice competition in theonline book industry.” The Journal of Industrial Economics, 50(3): 351–367.

Clemons, Eric K., Il-Horn Hann, and Lorin M. Hitt. 2002. “Price disper-sion and differentiation in online travel: An empirical investigation.” Manage-ment science, 48(4): 534–549.

CNN. 2000. “Amazon pricing flap - Sep. 28, 2000.” CNN Money.

Edelman, Benjamin. 2012. “Using Internet Data for Economic Research.” TheJournal of Economic Perspectives, 26(2): 189–206.

Ellison, Glenn, and Sara Fisher Ellison. 2009a. “Search, Obfuscation, andPrice Elasticities on the Internet.” Econometrica, 77(2): 427–452.

Ellison, Glenn, and Sara Fisher Ellison. 2009b. “Tax sensitivity and homestate preferences in internet purchasing.” American Economic Journal: Eco-nomic Policy, 53–71.

Euromonitor. 2014. “Internet vs Store-Based Shopping: The Global Move To-wards Omnichannel Retailing.” Euromonitor International.

FitzGerald, Drew. 2013. “Fear of ’Showrooming’ Fades — Best Buy, Other Re-tailers Optimistic That Price-Matching Can Stanch Trend.” Wall Street Jour-nal.

Gorodnichenko, Yuriy, and Oleksandr Talavera. 2016. “Price Setting inOnline Markets: Basic Facts, International Comparisons, and Cross-borderIntegration.” American Economic Review, Forthcoming.

Gorodnichenko, Yuriy, Viacheslav Sheremirov, and Oleksandr Talav-era. 2014. “Price Setting in Online Markets: Does IT Click?” NBER WorkingPaper Series, 20819.

Griffioen, Robert, Jan de Haan, and Leon Willenborg. 2014. “Collectingclothing data from the Internet.” Statistics Netherlands.

Horrigan, Michael. 2013. “Big Data: A Perspective from the BLS | AmstatNews.”

Kaplan, Greg, and Guido Menzio. 2015. “The morphology of price disper-sion.” International Economic Review, 56(4): 1165–1206.

Klena, Kali, and Jill Puleri. 2013. “From transactions to relationships: Con-necting with a transitioning shopper.” IBM Global Business Services.


Krsinich, Frances. 2015. “Price Indexes from online data using the fixed-effectswindow-splice (FEWS) index.” Statistics New Zealand.

Krsinich, Frances. 2016. “The FEWS index: Fixed effects with a windowsplice.” Journal of Official Statistics, 32(2): 375–404.

Lunnemann, Patrick, and Ladislav Wintr. 2011. “Price Stickiness in theUS and Europe Revisited: Evidence from Internet Prices*.” Oxford Bulletin ofEconomics and Statistics, 73(5): 593–621.

Mattioli, Dana. 2012. “On Orbitz, Mac Users Steered to Pricier Hotels.” WallStreet Journal.

Mikians, Jakub, Lszl Gyarmati, Vijay Erramilli, and NikolaosLaoutaris. 2012. “Detecting price and search discrimination on the internet.”79–84. ACM.

Mikians, Jakub, Lszl Gyarmati, Vijay Erramilli, and NikolaosLaoutaris. 2013. “Crowd-assisted search for price discrimination in e-commerce: First results.” 1–6. ACM.

Nakamura, Emi, and Jon Steinsson. 2008. “Five Facts About Prices:A Reevaluation of Menu Cost Models.” Quarterly Journal of Economics,123(4): 1415–1464.

Nygaard, Ragnhild. 2015. “The use of online prices in the Norwegian ConsumerPrice Index.” Statistics Norway.

Quan, Thomas W., and Williams. Kevin R. 2014. “Product Variety, Across-Market Demand Heterogeneity, and the Value of Online Retail.” UnpublishedPaper.

Simonovska, Ina. 2015. “Income differences and prices of tradables: Insightsfrom an online retailer.” The Review of Economic Studies, 82(4): 1612–1656.

Tang, Fang-Fang, and Xiaolin Xing. 2001. “Will the growth of multi-channelretailing diminish the pricing efficiency of the web?” Journal of Retailing,77(3): 319–333.

Valentino-DeVries, Jennifer, Jeremy Singer-Vine, and Ashkan Soltani.2012. “Websites Vary Prices, Deals Based on Users’ Information.” Wall StreetJournal.

Xing, Xiaolin, Zhenlin Yang, and Fang-Fang Tang. 2006. “A comparisonof time-varying online price and price dispersion between multichannel anddotcom DVD retailers.” Journal of Interactive Marketing, 20(2): 3–20.

Date post:	05-Jul-2020
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

Are Online and O ine Prices Similar? Evidence from Large Multi … · 2019-01-17 · Are Online and...

Documents